Home · Technology · Oct 29 archive
DeepSeek and Moonshot AI Use OpenAI's GPT-6 Astra for Model Training
Confirmed
In Short: DeepSeek and Moonshot AI, two prominent Chinese artificial intelligence laboratories, have trained their latest models using OpenAI's GPT-6 Astra, as reported by The Information and confirmed by Reuters.
Moonshot AI's Kimi K3.1 Pro model was launched last week, while DeepSeek's V4.1-Flash 1017 model was introduced on October 17, 2026.
OpenAI's GPT-6 Astra was first unveiled on April 9, 2026, and was made available to a limited set of organizations.
The Information, a technology industry-focused subscription outlet, originally reported this development.
This report marks a significant technical accusation in the ongoing AI landscape, as it names specific models, dates, and training methods.
The Information's report adds a new layer to the existing backdrop of China's push for open-weight AI and the U.S. Department of Homeland Security's order on DeepSeek.
For developers, buyers, and regulators, this story highlights the complex interplay between international AI development and regulatory oversight.
OpenAI previously announced it would no longer train its AI models on live internet data due to significant risks, a move that could affect how models like GPT-6 Astra are utilized.
Background
OpenAI said Friday it will no longer train its AI models on the live internet, citing significant risks.
What's confirmed
- Moonshot AI's Kimi K3.1 Pro model was launched last week, while DeepSeek's V4.1-Flash 1017 model was introduced on October 17, 2026.
- OpenAI's GPT-6 Astra was first unveiled on April 9, 2026, and was made available to a limited set of organizations.
- The Information, a technology industry-focused subscription outlet, originally reported this development.
- This report marks a significant technical accusation in the ongoing AI landscape, as it names specific models, dates, and training methods.
- The Information's report adds a new layer to the existing backdrop of China's push for open-weight AI and the U.S. Department of Homeland Security's order on DeepSeek.
- For developers, buyers, and regulators, this story highlights the complex interplay between international AI development and regulatory oversight.
- OpenAI previously announced it would no longer train its AI models on live internet data due to significant risks, a move that could affect how models like GPT-6 Astra are utilized.
What's still developing
- The original reporting came from The Information, a subscription outlet that covers the technology industry.
- Two of China’s best-known AI labs trained their newest models on the outputs of an American rival.
- Our earlier piece on China’s open-weight AI push and the DHS order on DeepSeek covered the policy side.
- That is the core claim in a report from The Information, relayed by Reuters on October 28, 2026.